Zomato AI Chatbot – Insights That Matter

 Industry Context: The GenAI Personalization Wave

This project is part of a teardown series examining how leading consumer tech brands apply GenAI to deliver contextual, personalized, and business-impacting experiences.

*Hypothetical context with Zomato, utilising the AI is the new UX wave

Company	Application Area	AI Usage	Impact
Zomato	Food delivery & support	GenAI chatbot (Together AI, Firefly orchestrator)	Can "double" customer satisfaction; support cost reduction
Google	Shopping	Virtual try-ons & AI-personalized search results	Improved UX, increased conversions, reduced decision fatigue
Amazon	E-commerce	GenAI for product Q&A, summaries, personalized listings	Shorter purchase cycles, better product discovery
Meta	Social & messaging	AI agents in Messenger, WhatsApp	1:1 conversational commerce, content suggestions
Duolingo	EdTech	GPT-4 for grammar feedback and chatbot tutors	Higher retention, increased engagement

 Strategic Insight

• AI is becoming the new UX layer. From shopping to food to education, users are being served in-context, persona-driven, and dynamically generated experiences.

What was once backend intelligence can be conversationally accessible!

• Zomato’s implementation highlights how GenAI can be applied in a practical, secure, and user-centric way — and forms the basis of our teardown series.

 Overview

Zomato has introduced a generative AI-powered chatbot using Microsoft Azure OpenAI and Together AI's Llama models, orchestrated internally through a service called Firefly. While it currently handles food ordering and support queries, the next frontier is enabling user-specific analytics like:

• "How much did I spend this month?"

• "How many times did I order biryani?"

This project outlines a solution to provide secure, personalized insights to users via the chatbot without live HSM decryption, ensuring speed, privacy, and scalability.

Architecture and Flow

1. User Asks a Query: "How much did I spend this month?"

2. Firefly Identifies the Intent: Classifies it as an analytics-type request.

3. Tokenized User ID Sent to Analytics API: No PII, only session-linked hashed ID.

4. Analytics API Fetches Precomputed Data: From Redis or DynamoDB.

5. Prompt Enriched: "User xyz has spent ₹2,350 and ordered biryani 5 times."

6. LLM Generates a Friendly Response: "You've spent ₹2,350 this month. Want to reorder your favorite biryani?"

7. Response Displayed: Through the chatbot in the Zomato app.

   
   
   
   

   

   
   

 Precomputed User Analytics Store

Field Name	Description
user_id_hash	Hashed or tokenized user ID
monthly_spend	Total amount spent this month
monthly_order_count	Number of orders this month
biryani_order_count	Number of biryani orders
most_frequent_item	e.g., “Chicken Biryani”
favorite_restaurant	Restaurant with most orders
total_order_count	All-time order count
avg_order_value	Average spend per order
loyalty_status	e.g., “Gold” or “Returning Customer”
last_order_item	Most recent dish

Psychological & Product Reasoning

Why Users Want It

• Personalization builds emotional connection.

• Budgeting convenience for students and working professionals.

• Gamified engagement through fun stats and progress.

• Ease of recall for frequent users (“What did I order last week?”).

Why Zomato Wants It

• Boosts retention and engagement.

• Creates cross-sell/upsell opportunities.

• Provides data-driven insight into user behavior.

• Adds a layer of AI differentiation in the market.

Why Users Might Not Want It

• Privacy and data tracking anxiety.

• Guilt over high spending if shown unprompted.

• Information overload for casual users.

Why Zomato Might Not Want It

• Infra cost for data orchestration and secure APIs.

• Risk of hallucination by LLMs damaging trust.

• Compliance risk under laws like India’s DPDP Act.

 PM Recommendation Matrix

Feature	Good For	Avoid For
Spend Recap (on request)	Budget users	Privacy-sensitive (Priya)
Favorite Dish/Frequency	Foodies	Casual users (Neha)
Gamified Tags & Badges	Loyalists	First-time users
Monthly Email Summary	Corporate users	Real-time engagement users

 Hook Canvas (Behavioral Design)

Element	Detail
Trigger	User sees chatbot prompt or asks about spend/order stats
Action	Clicks into chatbot and types or taps "How much did I spend?"
Variable Reward	Personalized response like "You’re in the top 5% biryani lovers"
Investment	User starts returning monthly to check stats, shares on social
Retention Loop	Continual engagement via monthly nudges and gamified feedback

User Persona

Persona	Use Case	Analytics Desire	Messaging Style
Ritika	Budget-conscious student	✅ Wants budget info	Supportive/opt-in
Anand	Power user foodie	✅ Loves stats	Fun, proactive
Neha	Casual explorer	❌ Not interested	None
Raj	Corporate spender	✅ Wants reports	Formal, email-based
Priya	Privacy-focused user	❌ Never	Full opt-out

 Go-To-Market Strategy

Phase 1: Alpha

• Internal employees + power users

• Monitor interaction quality, hallucination rate

Phase 2: Beta

• Early access to top 5% users

• Prompted via chatbot: "Want to see your foodie stats?"

Phase 3: Public

• Feature toggle for all users (opt-in)

• Launched via in-app banner + chatbot prompts

Messaging Strategy

• Chatbot: "You've spent ₹4,800 and ordered biryani 4 times. Want to reorder?"

• Email: "Your May food summary is ready."

• In-App Banner: "Track your foodie journey with Zomato AI."

KPIs

Metric	Target
Opt-in rate	≥ 25%
Repeat analytics usage	≥ 2x/mo
Drop-off post chatbot insight	< 10%
Chat satisfaction	≥ 90%
Negative feedback	< 5%